Cognex uses VisionPro® deep learning to provide deep learning-based image analysis software specifically designed for factory automation.
Traditional machine vision systems are usually the right choice for industrial image processing tasks that can be solved by developing rules. However, these systems often reach their limits when the objects they inspect vary and cannot be easily automated using rule-based programming. This issue occurs in food inspection, evaluation of soldered or welded seams, and many other applications. In these use cases, machine vision systems based on deep learning technology have recently established themselves as innovative solutions.
In the past, applying deep learning techniques required extensive knowledge of technology. Cognex VisionPro Deep Learning significantly reduces this entry barrier through a graphical user interface that simplifies training neural networks for users.
Intuitive graphical training
VisionPro deep learning tools are trained on images, unlike traditional image processing methods programmed with rule-based algorithms. VisionPro Deep Learning’s intuitive graphical user interface provides a simple environment for controlling and developing applications, significantly increasing the effort required for image collection, neural network training, and testing with different image sets. To reduce.
The Blue Locate tool is particularly suitable for automatic assembly or integrity checking, as it can accurately identify different parts.
Users can choose from four VisionPro deep learning image analysis tools. These are optimized for visual inspection in factory automation applications, so very few images are needed for rapid training. Blue Locate, Red Analyze, Green Classify, and Blue Read tools solve applications that are too complex for traditional rule-based image processing approaches.
Even if the background or surface texture changes, the RedAnalyze tool will find the smallest flaws. In the example above, the tool detects textile seam issues.
Identification by blue locate
The Blue Locate tool is ideal for tasks that need to detect or count parts that look different. Due to its rugged design, Blue Locate successfully identifies confusing backgrounds, low-contrast parts, and even parts that are bent, reshaped, or poorly lit. BlueLocate reliably identifies parts learned from sample images, regardless of perspective, orientation, brightness, gloss, or color. For these reasons, Blue Locate is, among other tasks, well suited for use in automated assembly validation.
Detect asylum with Red Analyze
If your parts have different background or surface textures, but you still need to find the smallest defects, RedAnalyze’s defect detection and segmentation tools are the right choice. By training good and bad part examples, you can tolerate normal deviations in terms of appearance while accurately detecting defects, impurities and other defects.
You can also use Red Analyze to segment variable regions in the image. Examples include welded, glued or painted areas, and background features that are dynamically hidden from the image to facilitate other inspections.
The Green Classify tool is a robust classifier that distinguishes between different object types, identifies error types, and inspects and classifies images.
Classification by green classification
VisionPro’s deep learning tool, Green Classify, is a robust classifier that can distinguish between different classes of objects, identify defect types, and classify good and bad parts. Green Classify is trained on a set of labeled images to identify objects based on common features such as color, texture, material, packaging, defect type, and classify them into classes. In doing so, the tool allows natural changes within the same class and effectively distinguishes acceptable changes from different classes. Green Classify handles complex classification tasks very quickly and does not require complex and time-consuming programming.
Optical character recognition with blue lead
Font and code reading and recognition is one of the most common tasks in image processing systems. Deformed, distorted, or poorly etched characters can cause real challenges and are easy to learn using the BlueRead tool. This tool uses a pre-trained deep learning font library to decipher difficult characters based on it.
The Blue Read tool reliably decodes and recognizes even deformed, distorted, or poorly etched fonts and code.
The user-friendly GUI eliminates the need for complicated programming and significantly reduces development time. The user simply defines the target area, sets the font size, and marks the text in the image. In just a few steps, you can train a robust tool to read application-specific plain text that traditional OCR tools cannot decode. In addition, training can be completed without any knowledge of image processing or deep learning. In addition, the optical debug function can detect misread characters and easily correct them.
Deep Learning and Rule-Based Vision: Best of Both Worlds
One of the main benefits of VisionPro deep learning is the unique ability to combine the available tools. This allows you to divide complex problems into smaller individual steps, simplify project optimization, and reduce the number of training images required. Another special feature of the new software environment is that developers can easily choose the best tool for a particular task by combining it with VisionPro’s rule-based vision library. Projects can be created in the VisionPro deep learning development environment and then exported to VisionPro projects. In this way, VisionPro and VisionPro Deep Learning offer the possibility to combine the best of both worlds.
https://engineering-update.co.uk/2021/09/16/simplify-neural-network-training-with-visionpro-deep-learnings-graphical-user-interface-and-intuitive-programming-environment/?utm_source=rss&utm_medium=rss&utm_campaign=simplify-neural-network-training-with-visionpro-deep-learnings-graphical-user-interface-and-intuitive-programming-environment Simplify neural network training with VisionPro deep learning graphical user interface and intuitive programming environment – Engineering Update